Calling conform and then unform with a spec that consists of some cat nested inside of some ? creates an extra level of nesting in the result:
This behavior does not occur with just ? or cat alone:
Patch: CLJ-2003-corrected.patch
This seems to be all that is needed.
The problem is actually more universal than (? (cat ...)). s/unform of a s/? with any regex child op will introduce an extra level of nesting. When the child is a regex, we are consuming the same "level" of sequence so unform should not introduce an extra level. However in other cases (non-regex ops), we should still possibly produce a nested collection.
The previous patch was too aggressive: it unwrapped all sub-unforms of s/?. This patch CLJ-2003-corrected.patch only unwraps when the sub-op is a regex.
Unfortunately it is impossible to distinguish between a desired-but-optional nil and a non-match from s/?. Specifically, the following tests now hold:
(I did not add these tests to the patch because I was unsure if they should be part of the contract of unform. However, they are pretty big gotchas.)
I also added tests for every possible subop of s/?, except ::s/accept, which I could not think of a test case for. (I'm not sure ::s/accept is actually reachable inside s/op-unform?)
Thanks for working on this - I will take a look when I get a chance.
I had to amend my patch slightly (same name): one of the test cases wasn't testing the correct thing.
Patch was no longer applying cleanly to master because of tests added by other commits. Patch rebased to master